Analysis of gene mutations, TMB, and PD-L1 in relation to ICI response in HNSCC.

Abstract

6027 Background: Immune checkpoint inhibitors (ICI) have changed the treatment of incurable advanced, recurrent, or metastatic (R/M) head and neck squamous cell carcinoma (HNSCC). Yet only a subset of patients respond, and predicting response remains challenging. We analyzed tumor genomic profiles and PD-L1 expression in a retrospective cohort of R/M HNSCC patients treated with ICI to evaluate the association with treatment response. Methods: We evaluated a single-institution cohort of 119 patients with R/M HNSCC treated with PD-1 inhibitors as monotherapy or in combination with palliative radiotherapy or chemotherapy, and who had genomics tested in tumor (tDNA) and/or in blood (ctDNA) samples. We analyzed clinical characteristics, treatment outcomes, PD-L1 expression, tumor mutation burden (TMB), and genomic profiles. Treatment response was assessed using iRECIST criteria. Responders were defined as complete response (CR), partial response (PR) ≥ 6 months, or stable disease (SD) ≥1 year. TMB and tDNA were tested by FoundationOne and ctDNA by Guardant. PD-L1 IHC used DAKO 22C3 antibodies. Results: Of 119 patients, 43 (36.1%) were considered responders (24 CR, 11 PR, and 8 SD ≥ 1 year). 97 patients had tDNA and TMB and 93 patients had ctDNA tested. 91 patients had PD-L1 results. In binary analysis, TMB ≥ 10 was significantly associated with response (P = 2.15e-5), while PD-L1 ≥ 20 was not (P = 0.83). A combined analysis of tDNA and/or ctDNA evaluated 273 genes. Univariate analysis showed that mutations in DNMT3A (P = 0.005), RET (P = 0.021), FAM123B (P = 0.021), and KDM6A (P = 0.043) were significantly associated with response. In a Lasso Logistic Regression model of the 27 most frequently mutated genes, 7 genes were significantly mutated in responders: DNMT3A (P = 0.0001), MAP2K4 (P = 0.025), FANCA (P = 0.036), ASXL1 (P = 0.019), EGFR (P = 0.0008), STK11 (P = 0.022), and BARD1 (P = 0.042), while TP53 was significantly mutated in non-responders (P = 0.010). Adding TMB to the multivariate model retained the significant association of DNMT3A (P = 0.021) and MAP2K4 (P = 0.015) mutations with response, and included ERBB4 (P = 0.037) mutations and TMB (P = 0.001) as additional significant indicators of response. DNMT3A was mutated only in responders (4 CR and 1 PR), with a positive predictive value of 1.00 and a negative predictive value of 0.74. In the multivariate analyses for tDNA and ctDNA, DNMT3A remained the only gene mutation significantly associated with response (P = 0.004 and P = 0.012). Conclusions: In this cohort of 119 patients with R/M HNSCC treated with PD-1 inhibitors, TMB, but not PD-L1, was associated with treatment response in univariate and multivariate analysis. Multivariate models identified eight genes significantly associated with treatment response. DNMT3A was the most remarkable gene, consistently associated with response in univariate and multivariate models. Further, larger studies are needed to validate these findings.